基于LSTM的飞行数据挖掘模型构建方法

提出了一种基于LSTM （Long Short Time Memory）模型的飞行历史数据挖掘模型的构建方法，此模型可以将飞行数据中有价值的目标数据自动提取出来。首先，通过滑动窗口法获得待检测数据；然后，将预先做好的训练样本数据输入到所构造的LSTM模型中进行训练，得到数据挖掘模型；最后，将待检测数据导入到训练好的LSTM模型中进行模式识别，将目标数据片段挖掘出来。结果表明，基于LSTM模型的飞行数据挖掘模型构建方法通用化程度高，可用于挖掘不同类型的目标数据，且识别率高，具有很高的工程应用价值。     

A tip clearance prediction model for multistage rotors and stators in aero-engines

Tip clearances of multistage rotors and stators greatly affect aero-engines’ aerodynamic efficiency, stability and safety. The inevitable machining and assembly errors, as well as the complicated error propagation mechanism, cause overproof or non-uniform tip clearances. However, it is generally accepted that tip clearances are difficult to predict, even under assembly state. In this paper, a tip clearance prediction model is proposed based on measured error data. Some 3D error propagation sub-models, regarding rotors, supports and casings, are built and combined. The complex error coupling relationship is uncovered using mathematical methods. Rotor and stator tip clearances are predicted and analyzed in different phase angles. The maximum, minimum and average tip clearances can be calculated. The proposed model is implemented by a computer program, and a case study illustrates its performance and verifies its feasibility. The results can be referred by engineers in assembly quality judgement and decision-making.     

飞机起落架全方位移动装配机器人设计与研究

起落架作为支撑飞机重量、吸收撞击能量的重要部件,在装配和试验过程中面临着装配空间狭小、装配精度高等特殊要求,急需研制一款具备可移动、多自由度调节功能的柔性装配平台。通过对起落架装配需求进行分析,设计出基于Mecanum轮的全方位移动和基于3–RPS的并联机构组成的6自由度装配机器人,详细介绍了装配机器人的结构组成、控制方法,对机器人运动学特性进行分析获得运动学模型,并进行运动学参数设置及精度分析。通过对起落架装配机器人精度测试以及实物装配,验证该机器人的功能和性能满足使用要求。     

Parameter studies and evaluation principles of delamination damage in laminated composites

Delamination represents one of the most severe failure modes in composite laminates, especially when they are subjected to uniaxial compression loads. The evaluation of the delamination damage has always been an essential issue of composite laminates for durability and damage tolerance in engineering practice. Focusing on the most typical and representative elliptical delamination issue, an analytical model simultaneously considering the conservative buckling process and non-conservative delamination propagation process is implemented. Various computational cases considering different delamination depths, directions, aspect ratios, and areas are established, and the predicted results based on the analytical model are carefully compared. Effects of these geometrical delamination parameters on the buckling, delamination propagation, and failure behaviors of composite laminates are thoroughly analyzed, and innovative evaluation principles of the delamination damage have been concluded. It is found that the delamination area is the key factor that truly affecting the failure behaviors of delaminated composites, and the local / global buckling and failure loads show clear linearity with the delamination area, whilst the delamination depth and direction only have slight effects.     

基于Deform软件二次开发和BP神经网络的TA15多向锻造微观组织预报

通过热压缩实验构建了TA15合金的热变形真应力-应变曲线，以此为基础分别建立了合金双相区及单相区温度区间的热变形本构方程；基于热压缩试样动态再结晶的统计数据建立了TA15合金的动态再结晶模型。借助Deform提供的二次开发功能实现相关数学模型的程序化，制定正交实验方案，实现了TA15合金多向锻造变形的微观组织仿真。通过正交实验分析得出各项因子的影响对象及强弱差异，提出了双相区及单相区温度区间内的多向锻造最佳因子组合。建立了TA15合金多向锻造变形微观组织的BP （Back Propagation）神经网络预测模型，将预测结果与有限元仿真结果进行比较，结果表明两种方法的预测结果基本一致，但神经网络具备有限元仿真难以实现的良好细节预测能力，能更为细致地实现对微观组织分布状态的划分。     

柔性支撑控制力矩陀螺动力学建模与控制

针对柔性支撑引起控制力矩陀螺(CMG)框架控制性能下降的问题，建立了柔性支撑条件下控制力矩陀螺的动力学模型，分析了柔性支撑扰动力矩的耦合机理并给出了解析模型.据此，设计了一种积分滑模控制策略，用于提升对CMG框架动力学特性变化和新增扰动力矩的鲁棒性.仿真结果表明，该控制策略相比传统的PID控制具有更好的动态过程和稳态性能，提升了柔性支撑下控制力矩陀螺框架速度的闭环控制性能.     

基于机器学习的空间翻滚目标实时运动预测

借助监督式机器学习（ML）方法，对空间翻滚目标的运动状态预测问题进行研究，为空间机器人抓捕空间翻滚目标提供可靠的数据依据。基于物理模型的运动预测方法依赖理想的建模假设，需要连续的视觉反馈信息，解决目标预测问题的能力有限。因此，本文采用机器学习中纯数据驱动方式的稀疏伪输入高斯过程（SPGP）回归方法进行空间翻滚目标的运动预测。给定空间翻滚目标运动状态的历史观测数据，通过连续优化真实观测数据，得到稀疏的伪训练数据集，进而在线快速预测目标的运动状态，预测的计算效率达到毫秒级。此外，利用马尔科夫链蒙特卡洛（MCMC）法处理连续优化过程，克服由于随机初始值造成的优化过程陷入局部极小值问题。利用Snelson数据验证了所提稀疏伪输入高斯过程回归方法的正确性，并通过4组仿真算例验证了所提方法对于空间翻滚目标运动预测的有效性和鲁棒性。     

面向集中力扩散的回转曲面加筋拓扑优化方法

为了提高航天器回转曲面加筋型连接结构的集中力扩散效率，需要开展回转曲面加筋集中力扩散结构设计。传统放射肋设计方法普遍依赖设计经验、难以满足集中力高效扩散需求。因此，提出一种面向集中力扩散的回转曲面加筋拓扑优化方法。第1步，建立了一种基于各向异性过滤技术的集中力扩散拓扑优化方法，保证拓扑优化结果满足回转曲面加筋制造工艺要求；第2步，提出了一种基于网格变形技术的拓扑优化结果智能重构方法，可高效准确地对回转曲面加筋拓扑优化结果进行模型自动重构。基于所提出方法，以卫星平台对接环这种典型的回转曲面加筋壳为对象开展算例研究，并将优化结果与传统放射肋设计结果进行对比。结果表明，所提出的优化方法可得到加筋构型清晰、满足回转曲面加筋制造工艺要求的优化结果，且具有集中力扩散效率高、网格质量依赖性低、拓扑特征重构高效等优点。     

Theoretical modeling of vectoring dual synthetic jet based on regression analysis

The excellent vectoring characteristic of Dual Synthetic Jet (DSJ) provides a new control strategy for the active flow control, such as thrust vectoring control, large area cooling, separated flow control and so on. For incompressible flow, the influence relation of source variables, such as structure parameters of actuators, driving parameters and material attributes of piezoelectric vibrating diaphragm, on the vectoring DSJ and a theoretical model are established based on theoretical and regression analysis, which are all verified by numerical simulations. The two synthetic jets can be deemed as a main flow with a higher jet velocity and a disturbing flow with a lower jet velocity. The results indicate that the influence factors contain the low-pressure area formed at the exit of the disturbing flow, which could promote the vectoring deflection, and the impact effect of the disturbing flow and the suppressive effect of the main flow with the effect of restraining the vectoring deflection. The vectoring angle is a complex parameter coupled by all source variables. The detailed theoretical model, whose error is controlled within 3.6 degrees, can be used to quantitatively assess the vectoring feature of DSJ and thus to provide a guidance for designing the control law applied in the active flow control.     

A novel virtual material layer model for predicting natural frequencies of composite bolted joints

A novel virtual material layer model based on the fractal theory was proposed to predict the natural frequencies of carbon fiber reinforced plastic composite bolted joints. Rough contact surfaces of composite bolted joints are modeled with this new proposed approach. Numerical and experimental modal analyses were conducted to validate the effectiveness of the proposed model. A good consistence is noted between the numerical and experimental results. To demonstrate the necessity of accurately modeling the rough contact surfaces in the prediction of natural frequencies, virtual material layer model was compared with the widely used traditional model based on the Master-Slave contact algorithm and experiments, respectively. Results show that the proposed model has a better agreement with experiments than the widely used traditional model (the prediction accuracy is raised by 8.77% when the pre-tightening torque is 0.5 N∙m). Real contact area ratio A* of three different virtual material layers were calculated. Value of A* were discussed with dimensionless load P*, fractal dimension D and fractal roughness G. This work provides a new efficient way for accurately modeling the rough contact surfaces and predicting the natural frequencies of composite bolted joints, which can be used to help engineers in the dynamic design of composite materials.